BlackPrint Technologies · Aug 2026
GeoKG Lab
Governed Geospatial Knowledge Graph with Claude Tools
Overview
Researched and built a reproducible geospatial knowledge-graph demo that keeps the spatial database as the source of truth while serving a validated read-only snapshot to graph queries, an expansion lab and optional Claude tool use.
Highlights
- 01
Built a read-only extraction path (allowlisted SELECT queries, read-only transactions and rollback) into a content-hashed snapshot of 16 municipalities and 128 H3 cells, described by an OWL vocabulary with SHACL constraints.
- 02
Implemented an Expansion Lab with Pareto shortlists, sensitivity checks and portfolio sequencing, plus multi-hop graph queries that return evidence chains with explicit caveats that adjacency is not a route.
- 03
Gated an optional Claude Haiku 4.5 conversation behind explicit opt-in, typed local tools, numeric grounding checks and per-turn and per-session cost caps, verified by 149 backend and 31 frontend tests.
Stack
- Python
- Flask
- NetworkX
- PostgreSQL (read-only)
- OWL / Turtle
- SHACL
- GeoSPARQL
- H3
- Claude Haiku 4.5
- Tool Use
- Pareto Analysis
- Knowledge Graphs
Working on something similar?
Tell me about the problem and the data behind it. I reply within 24–48 hours.
Discuss a project